August 10, 2026
Wiring Up CodeRabbit — a Second Reviewer for Every Change
Connected CodeRabbit's GitHub App and CLI to this repo, then proved both actually work by deliberately breaking a script and watching the CLI catch it before it was ever committed.
2
integrations verified (GitHub App + CLI)
9/9
CLI environment checks passing
1 major
real bug caught in a live test, then reverted
Problem
Every change to this site — mine or one of the site's own agents' — ships straight to main with no second reviewer. As the number of automated write paths grew (Telegram agent, MCP tools, scheduled agents), a low-effort second set of eyes on the actual diff became worth having, especially for changes an agent makes without a human reading every line first.
Approach
Installed the CodeRabbit GitHub App (already reviewing pull requests) and the CodeRabbit CLI (`brew install --cask coderabbit`) for local, pre-commit review. Rather than trust the setup on faith, ran a real test: deliberately removed the regex-escaping in scripts/check-tailwind-classes.mjs — a subtle bug where a Tailwind class name gets interpolated straight into a RegExp without escaping metacharacters — then ran `coderabbit review --agent` against the uncommitted diff.
Outcome
CodeRabbit caught it immediately: a correctly-flagged major-severity finding, pinpointing the exact missing escape step and suggesting the right fix. The change was reverted right after — it was never committed, since the point was proving the tool actually reads and understands the code, not just that it runs. Both the GitHub App (PR-level) and CLI (pre-commit) are now part of the normal workflow: review before committing, second review on the PR.